Estimation of harvested fruit weight using volume measurements with distance sensors: A case study with olives in a big box

https://doi.org/10.1016/j.compag.2023.107620Get rights and content
Under a Creative Commons license
open access

Highlights

  • Two Time-of-Flight technologies have been tested to develop a yield monitor.

  • Two algorithms have been developed for weight estimation by volume measurement.

  • Some factors such as illumination or acceleration affect the accuracy.

  • The developed systems are more stable than load cells under field conditions.

  • The systems developed are a valid alternative to weighing with load cells.

Abstract

The introduction of fruit tree harvesters on the market allows the opportunity to incorporate new yield monitors. Traditionally, yield has been measured by in-the-field fruit batch weighing systems that employ load cells, which present certain problems as well as oscillations and outliers. We propose the development and evaluation of two distance measurement systems (an experimental one using an array of sensors with low-cost, Time-of-Flight technology and another, commercial system, with a 3D camera) to estimate the volume of fruit harvested in a big box and correlate the volume with its weight. To this end, two algorithms were developed to estimate the volume of fruit filling. Several tests were conducted to determine the field of view of the sensors and the influence of illumination, reflectivity of different surfaces, and vibrations in transit on the measurements they give. Illumination was a limiting factor on the accuracy of the experimental system and required mitigating actions to operate with it. The mean relative error of sensor distance measurement was less than 0.8 % and 1.6 % for the commercial and experimental systems, respectively, which decreased as distance from the measurement target increased. Measurements on matte surfaces showed a lower measurement error than those on glossy surfaces, with error being twice as high in the commercial system than in the experimental system. The error in volume estimation was lower in the commercial system and could be reduced to less than 1.6 % with pre-calibration. In general, with the accelerations typical of agricultural traffic, in dynamic operation the distance sensors provided less variation in results than the load cells, which would require processing of the recorded signals. In the range of filling a box over 150 kg, the absolute error in weight estimation was 5.4 kg for the experimental system and 11.0 kg for the commercial system. However, this error may increase with the use of the experimental system if filling occurs from the center or from the corner. In general, the systems offer acceptable results for using this technology if extreme accuracy is not required. This work establishes the basics of a technology that can be an alternative to load cells and be applied to harvesting machinery to record continuous real-time production.

Keywords

ToF
3D camera
Load cells
Yield monitor
Mechanization

Data availability

No data was used for the research described in the article.

Cited by (0)